Can Transcranial Direct-Current Stimulation Alone or Combined With Cognitive Training Be Used as a Clinical Intervention to Improve Cognitive Functioning in Persons With Mild Cognitive Impairment and Dementia? A Systematic Review and Meta-Analysis

Background: Transcranial direct-current stimulation (tDCS) facilitates cognitive improvement in healthy and pathological populations. It has been increasingly used in cases of mild cognitive impairment (MCI) and dementia. Our research question is: Can tDCS serve as a clinical intervention for improving the cognitive functions of persons with MCI (PwMCI) and dementia (PwD)? Objective: This systematic review evaluated the evidence to determine the efficacy of tDCS in improving cognitive outcomes in PwD and PwMCI. Methods: A systematic review was conducted of studies published up to November 2017 involving tDCS in cases of MCI and dementia. Studies were ranked according to the level of evidence (Oxford Center for Evidence-Based Medicine) and assessed for methodological quality (Risk of Bias Tool in the Cochrane Handbook for Systematic Reviews of Interventions). Data was extracted on all protocol variables to establish a reference framework for clinical interventions. Different modalities, tDCS alone or combined with cognitive training, compared with sham tDCS were examined in both short and long-term effects. Four randomized control trials (RCTs) with memory outcomes were pooled using the fixed-effect model for the meta-analysis. Results: Twelve studies with 195 PwD and four with 53 PwMCI met the inclusion criteria. Eleven articles were ranked as Level 1b. The results on the meta-analysis on pooled effects of memory indicated a statistically significant medium effect size of 0.39 (p = 0.04) for immediate effects. This improvement was not maintained in the long term 0.15 (p = 0.44). Conclusion: tDCS improves memory in PwD in the short term, it also seems to have a mild positive effect on memory and language in PwMCI. However, there is no conclusive advantage in coupling tDCS with cognitive training. More rigorous evidence is needed to establish whether tDCS can serve as an evidence-based intervention for both populations.


INTRODUCTION
Transcranial direct-current stimulation (tDCS) is a type of non-invasive brain stimulation (NIBS). tDCS delivers weak direct currents to the brain that can alter spontaneous firing rates on neural activity, which subsequently translates into behavioral changes . It is a process that has been described as "portable, painless, inexpensive and safe" (Kadosh et al., 2012). During the administration of tDCS, depolarization or hyperpolarization of the neuronal membrane of target neurons may be induced, even though the small electric fields of tDCS are considered to be below the intensity required to evoke action potentials (Nitsche et al., 2003;Miniussi et al., 2013;Tatti et al., 2016). In other words, tDCS causes a shift in the membrane potential threshold which is likely to change the probability that an incoming action potential will result in post-synaptic firing during and after its administration (Prehn and Flöel, 2015). Such changes in neuronal excitability modulates the cognitive processes and tDCS can induce physiological processes. Due to the proposed resemblance of the effects of tDCS and cognitive processes on cerebral physiology, researchers have been using NIBS to alter cognition (Kuo and Nitsche, 2012;Prehn and Flöel, 2015).
Mild Cognitive Impairment (MCI) is defined as the stage between normal and dementia-type pathological aging. MCI is a syndrome of cognitive decline in non-demented persons that does not affect the capacity to be independent in activities of daily living (ADLs; Portet et al., 2006). In contrast, people who suffer from dementia present a more severe cognitive decline and do not preserve independence in functional abilities and ADLs (Langa and Levine, 2014). Epidemiological investigations suggest a range of prevalence for MCI of 7-24% among adults aged over 65, and the manifestation of MCI is consistently shown to have a high risk of progression to dementia (Langa and Levine, 2014;Petersen et al., 2014). To date, there is no pharmaceutical treatment shown to be effective in improving cognitive functioning in MCI and dementia (Langa and Levine, 2014), although cognitive training interventions show promise for improving targeted cognitive functions in elderly persons without cognitive impairments (Ball et al., 2002). Cognitive Rehabilitation (CR) is defined as "the therapeutic process of increasing or improving an individual's capacity to process and use incoming information so as to allow increased functioning in everyday life." This includes methods to train and restore cognitive functioning as well as compensatory techniques (Sohlberg and Mateer, 1989, p. 871).
CR is therefore essential and research has indicated that NIBS can positively affect the cognitive performance of populations affected by cognitive disorders (Miniussi et al., 2008). Differences in tDCS experimental protocols regarding the parameters employed such as the montage, the current, the intensity or the size of the electrodes can affect the electric field strength. All of these variables contribute to increase the heterogeneity of the electric field's properties among studies thus producing different outcomes (Woods et al., 2016). Furthermore, targeting a neural network with tDCS while it is engaged by a cognitive stimulation activity, during or after the administration of tDCS, may yield better therapeutic effects than stimulating the same cortical region lacking cognitive stimuli (Cruz Gonzalez et al., 2018). tDCS may increase the strength of transmission across synaptic circuits in pathways that are stimulated by cognitive practice. Thus, coupling both techniques could create a synergistic positive effect on behavior (Miniussi et al., 2013;Birba et al., 2017;Cruz Gonzalez et al., 2018). The effectiveness of tDCS in CR targeting people with MCI or dementia must therefore be established. It is fundamentally important to learn about all the different configurations and protocols in which tDCS has been employed to assess its utility.
We systematically reviewed the literature regarding effects of tDCS on persons with MCI and dementia to address the following questions: (1) Does tDCS alone improve cognitive functioning in persons with MCI and dementia? (2) Does tDCS coupled with cognitive training, or as a priming to other cognitive interventions yield greater benefits in cognitive functioning than the administration of tDCS alone? (3) Are the effects of tDCS on the cognitive functions able to maintain across time?
In this study, we reviewed and evaluated the effects of tDCS on cognitive functions in people with MCI or dementia from all the available clinical trials. A systematic review of the available information up to the present will enable researchers to better understand the potential of tDCS to offer solutions for cognitive deterioration, with the aim of outlining more robust interventions in the future for people with MCI and dementia. Other reviews involving the use of different NIBS on healthy aging (Prehn and Flöel, 2015), dementia (Freitas et al., 2011;Hsu et al., 2015), MCI (Birba et al., 2017) have been carried out since 2011, but we provide an update and meta-analysis of recent trials to focus exclusively on the use of tDCS in MCI and dementia populations.

Eligibility Criteria
We performed a systematic review and meta-analysis following the PRISMA guidelines (Liberati et al., 2009). Studies were selected based on the following criteria: -Participants: Participants included in the study were older adults with MCI and persons with a diagnosis of dementia. The criteria for MCI includes (a) subjective memory complaint; (b) objective cognitive decline; (c) preserved ADLs, and (d) not demented (Petersen et al., 1999). The diagnosis of dementia followed the criteria of the NINCDS-ADRDA (McKhann et al., 1984) and the DSM-IV (American Psychiatric Association, 2000). Participants with any other neurological disease that was not dementia, such as only the Parkinson's type, were excluded. -Interventions: tDCS alone (anodal, cathodal, or sham), or a combination of tDCS (online or offline) with an additional cognitive task (CT).

Selection Criteria
After removing duplicates, the abstracts of the articles retrieved were screened to make a final decision for further review. Two investigators realized the search and the selection of studies to be included. Any disagreements were resolved by a third reviewer.

Data Extraction
The data extracted from the selected studies were conducted by two investigators using a standardized data extraction sheet which included study design, study population, number of participants, mean participant age, gender ratio, general cognitive level, number of intervention sessions, experimental/sham tDCS parameters, combination of tDCS with other interventions, outcome measures, neuroimaging techniques, assessment sequence, follow-up, effect(s) of the intervention, and intervention safety reports.

Methodological Quality
The studies selected for review were categorized and leveled according to their design based on the hierarchy level of  trials (RCTs) were then rated by the first two authors using the Risk of Bias Tool in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins and Green, 2008).

Data Analysis
Only RCTs, excluding crossover designs, were considered for meta-analysis. In some cases, authors were contacted to obtain data from their studies. After the review of the clinical methodology's heterogeneity of each study (Table 2), the selected papers were further assessed for statistical heterogeneity, using the I-squared and Chi-squared statistics of the outcome measures. Data of pooled memory outcomes comparing: (1) Short-term effects of tDCS treatments vs. sham tDCS that targeted the dorsolateral prefrontal cortex (DLPFC) were calculated based on the differences between post-intervention evaluations relative to the baseline to assess the immediate effects of tDCS; (2) Longterm effects of tDCS treatments vs. sham tDCS that targeted the DLPFC; were assessed according to the differences between follow-up evaluations relative to the baseline.
All outcomes were analyzed as continuous variables with the mean change, the largest standard deviation, and the sample size in each group. The standardized mean difference and 95% confidence intervals were calculated for all meta-analyses using the fixed-effect model. The effect size was considered to be small between 0.2-0.49, moderate (0.5-0.79), and a value of 0.8 or above was considered to be large (Cohen, 1992). If I 2 was below 40%, it was considered to not represent statistical heterogeneity. Otherwise, the random-effect model was used instead. Significance was set at p = 0.05 and both meta-analyses were conducted using Review Manager Software 5.3.
Four studies (Boggio et al., 2012;Cotelli et al., 2014;Khedr et al., 2014;Bystad et al., 2016b) assessed the long-term effects of tDCS. Three of these reported significant changes: one showed that the improvement caused by anodal tDCS persisted 4 weeks after the end of stimulation (Boggio et al., 2012), another indicated that either anodal or cathodal tDCS improved mean MMSE score at 1-and 2-month follow-up (Khedr et al., 2014), and the third study revealed that 2 months after the end of the intervention, anodal tDCS was clinically significant (Bystad et al., 2016b).
Only two studies performed neuroimaging tests. In the first, an ERP experiment confirmed significant effects reducing P300 latency after both anodal and cathodal tDCS (Khedr et al., 2014). The second used EEG, although it did not prove changes from baseline (Bystad et al., 2016b).
of anodal tDCS. Yun et al. (2016) utilized PET to demonstrate a significantly increased metabolism in cortical regions. In the same way, the work of Ladenbauer et al. (2017) made clear, through the use of concurrent EEG, that slow oscillatory tDCS significantly increased overall slow oscillations (SO) and spindle power (Ladenbauer et al., 2017).

Effectiveness of tDCS Combined With CT
Details and methods about the CT operated among studies are shown in Table 3. All the studies involving PwD showed significant benefits after receipt of anodal tDCS paired with a CT. Boggio et al. (2009)applied tDCS while participants completed cognitive assessments, enhancing memory in a visual recognition memory task, but there were no effects on attention. The work of Cotelli et al. (2014) combining memory training with tDCS and sham tDCS resulted in improved memory performance illustrated in a face-name association memory task, as compared to a group which received tDCS paired with motor training; this improvement persisted significantly after 12 weeks. However, it failed to produce significant effects on standardized cognitive tests. In one single-subject crossover study, the cognitive training associated with memory components was started right after the end of tDCS administration and the findings revealed a significant accuracy improvement in a verbal working memory task. In contrast, there is no indication of amelioration in other cognitive assessments (Penolazzi et al., 2015). Alternatively, one case study that focused on stimulating the production and comprehension of language through a combination of anodal tDCS and linguistic training found a significant effect in an auditory comprehension task (Costa et al., 2017). The work of Meinzer et al. (2015) targeting PwMCI revealed that during exposure to anodal tDCS, participants performed significantly better in a semantic word-retrieval task than those who received sham tDCS, achieving the level of healthy elderly subjects. Furthermore, the application of anodal tDCS led to reduced task-related prefrontal hyperactivity shown by restingstate fMRI.

Study Quality
The level of evidence of all the trials is displayed in Figure 1. Details can be found in Table 4. Most of the studies reported a risk of bias describing the method used to conceal the allocation sequence (Ferrucci et al., 2008;Boggio et al., 2012;Meinzer et al., 2015;André et al., 2016;Bystad et al., 2016a;Yun et al., 2016;Ladenbauer et al., 2017). The most common methodological limitation of these studies was the issue of the blinding of the personnel due to the nature of most tDCS devices.

DISCUSSION
All the 11 articles (RCTs) whose evidence was ranked as level 1b presented a commendable methodological quality with a general presence of low risk of bias. From the MMSE admission scores in the AD studies that ranged from 15 to 24.5 and MCI studies from 26.75 to 28.3, we noticed that the effects of tDCS benefits on cognition were significantly better for patients with mild to moderate cognitive decline. When comparing the effectiveness of tDCS, in single and multisession interventions, positive changes occurred in both behavioral and neural systems. In this systematic review, we aimed to reveal robust interventions by identifying similar elements across studies. One main concern when designing interventions in NIBS is the treatment duration in multisession trials. There is similarity in terms of the number of sessions across the selected studies: four to ten sessions, staggered over 1-2 weeks. These short interventions can provide valuable data that allow tDCS to be proposed as a potential option in CR. However, the benefit is rather short-term with a medium effect size of 0.39. This also contrasts with other long intervention frameworks for clinical use in which more time is needed to evaluate whether the changes have a real benefit in reversible conditions such as MCI (Portet et al., 2006) or have an impact in long-term neurodegenerative processes such as dementia. For example, an alternative was proposed by Bystad et al. (2017) that adopted an 8-month protocol of daily tDCS use in a person with AD to stabilize cognitive decline. The long-term outcome probably requires prolonged periods of intervention.
Although six studies reported side effects (Ferrucci et al., 2008;Khedr et al., 2014;Suemoto et al., 2014;Bystad et al., 2017;Ladenbauer et al., 2017;Murugaraja et al., 2017), all participants tolerated the therapies well and the sensations experienced were mild. This suggests that the parameters employed are sufficiently safe (up to 30 min, 2 mA). Another concern is that the range of the parameters for intensity and duration stimulation and the size of the electrodes were highly diverse, making it difficult to draw conclusions in order to select a specific protocol for future research.
Another view is that when selecting a region of interest for stimulation, most of the studies targeted the temporal regions (Ferrucci et al., 2008;Boggio et al., 2012;Bystad et al., 2016aBystad et al., ,b, 2017, for the role this area plays in certain memory processes (Brown et al., 1987;Kaye et al., 1997) as well as language (Nguyen et al., 2018). Another common region of interest is the DLPFC because of its importance in high-order cognitive mechanisms (Tremblay et al., 2014). Language-oriented work has targeted the inferior frontal gyrus and DLPFC as well, successfully achieving better performance in semantic word retrieval (Meinzer et al., 2015) and comprehension of language (Costa et al., 2017). In the same way, studies that applied tDCS combined with CT operated a CT related with a cognitive domain associated with the brain area targeted by tDCS. Although this approach is reasonable and consistent, the studies failed to assess if other cognitive domains associated with other brain regions were affected. Due to the lack of focality of tDCS and the variability of the current flow direction, there is a possibility that other neural networks, not directly targeted by tDCS, could have been affected (Woods et al., 2016).
Overall, these studies have selected predominantly global cognition and memory domain as experimental evaluators. Despite the fact that these constructs are similar in nature, there is great variability in terms of assessment and CT chosen. All the studies but two (Suemoto et al., 2014;Bystad et al., 2016a) report positive effects of the application of tDCS. Against this trend, among the other articles, we must emphasize that only six studies translated these improvements into standardized cognitive assessments (Ferrucci et al., 2008;Khedr et al., 2014;André et al., 2016;Bystad et al., 2016b;Yun et al., 2016;Ladenbauer et al., 2017) while other studies reporting improvements in non-standardized CT to prove the effects of tDCS. Yet it must be acknowledged that certain cognitive functions are mediated by networks of various brain sites and might be difficult to be influenced by targeting only a subset of their brain regions (Reinhart et al., 2017), besides the short length of the intervention might have contributed to these changes being insufficient to translate into standardized test results.
It is hypothesized that targeting a neural circuit with tDCS paired with a CT may produce stronger therapeutic effects than stimulating the same brain area without cognitive stimuli (Birba et al., 2017;Cruz Gonzalez et al., 2018). The evidence on whether using tDCS alone or in combination with other CT yields identical results and seems to be inconclusive in both PwD or PwMCI. Recently, a single-subject design study using cognitive stimulation practice across sessions in combination with simultaneous anodal tDCS showed significantly stronger effects on planning ability, processing speed, and attention of cognitive stimulation practice than both sham tDCS and the application of cognitive stimulation practice alone in PwMCI (Cruz Gonzalez et al., 2018). This finding prompts the plausible speculation that tDCS, combined with cognitive training, might have synergic effects. A recent review of CR or cognitive training interventions with control conditions for PwD shows that RCTs on the effect of cognitive training on PwD are limited and there is no indication of any significant benefits from cognitive training (Bahar-Fuchs et al., 2013). Following this line of thought, future studies would carry more weight if they considered combining both interventions in comparison with control groups receiving tDCS or cognitive training alone, and would report not just benefits in the trained CT but also generalization to the trained cognitive domains and daily functioning.
Only five studies reported the use of brain imaging as an outcome demonstrating the neuromodulatory effects of tDCS (Khedr et al., 2014;Meinzer et al., 2015;Bystad et al., 2016b;Yun et al., 2016;Ladenbauer et al., 2017). In the absence of imaging techniques, we can only speculate on the results of behavioral tests without examining the underlying neural mechanism of tDCS in MCI or dementia. This is the first meta-analysis to explore the short-and longterm effects of tDCS in the memory domain, targeting the DLPFC in PwD. We have found evidence that tDCS has a significant immediate effect but that it is not significantly sustained with the passage of time. We suggest that future research address the need to evaluate the long-lasting effects of tDCS on the cognitive domain, implementing both behavioral and imaging follow-up evaluations.
This study has several limitations. For instance, although the pooled outcomes for meta-analysis were all memory-based, the selected studies used different tests. In addition, only four studies could be included, this might contribute to making the meta-analyses somewhat underpowered, thus the findings should be interpreted with cautions. Another striking example is the AD stage, which varied among the studies. Moreover, we have not included the most recent work published since November 2017 (Cruz Gonzalez et al., 2018), because of the time eligibility criteria. This systematic review included all tDCS trials carried out in dementia and MCI populations, and subsequently reported a few papers that did not use a comparison group (sham tDCS), which weakens the conclusions somewhat.

CONCLUSION
Our meta-analysis suggests that there is modest evidence supporting tDCS on the DLPFC ameliorates memory in PwD, however, the benefits are not long-term. Our review shows that tDCS alone seems to have a positive effect on cognition particularly for memory and language in PwD, with mild to moderate cognitive decline, and MCI. Whether tDCS might produce better outcomes on PwMCI and PwD in coupling with another CT than when administered alone remains unclear.
Although all these findings are promising, the administration of tDCS might not yet be a valid option for clinical intervention for dementia or MCI. Some of the results come from non-RCT studies, and the heterogeneity of the clinical trials does not allow one to define a clear protocol with optimal parameters. Furthermore, the interventions were too short to determine the real effects on cognitive functions and none of the studies assessed the impact of treatments on everyday cognition in daily functioning, which is an essential domain to be considered due to the functional consequences of dementia. We recommend that future studies include prolonged periods of intervention, neuroimaging techniques, and consider more robust, standardized methodology of tDCS in order to establish whether tDCS can serve as an evidence-based clinical intervention for PwMCI and PwD.